Algorithms and Structures of Neurocomputers

Lecturer:

Tutor:

Supervisor:

Department of Circuit Theory

Synopsis:

Information about the basic principles and possibility of the application of the neural informative technology for the signal processing are the main topic. The lectures are devoted to the introduction into the artificial neural networks (NN) theory and applications, to the choice and the optimisation of the structures and the neural network applications at the speech and image processing are investigated in detail. Some neural network applications in the biomedical engineering and hardware realization of the KSOM are described.

Requirements:

Basic knowledge of the speech and image processing, MATLAB, probability calculus and statistics applications. Active participation on the seminars, develop semester project. More on http://amber.feld.cvut.cz/SSC

Syllabus of lectures:

1. Neural networks - research history, biological and artificial NN, applications

for signal processing, neural models, activation functions.

2. Learning principles, Self-Organizing Maps (SOM), Kohonen's maps.

3. Supervised SOM, U-matrix, LVQ classifier.

4. Multilayer networks with Back-Propagation learning algorithm (BPG).

5. Basic BPG, modifications.

6. Optimisation of the structure, neural network pruning, data mining.

7. Basic terms of phonetics, characteristics of the speech.

8. Methods of the speech recognition, neural networks applications.

9. Principles of the speech synthesis, types of the synthesizers.

10. Artificial neural networks (ANN) for speech synthesis.

11. Artificial neural networks (ANN) in biomedical engineering.

12. Associative memory, Hopfield networks, ART networks.

13. The others ANN applications.

14. Special paradigms (CNN, TDNN, Wavelet networks, fuzzy-neural networks, GA).

Syllabus of tutorials:

1. Introduction, MATLAB, NN-Toolbox fundamentals, information of the semester

projects.

2. ANN basic function, Perceptron, ADALINE, MADALINE, LMS algorithm.

3. Self-Organizing Maps, supervised SOM, U-matrix. SOM Toolbox.

4. Kohonen's maps, LVQ algorithms - NN Toolbox, MATLAB.

5. Multilayer neural networks. Assignment of the semester projects.

6. Modifications of the BPG algorithm.

7. Speech Laboratory - experiments.

8. SOM Laboratory - experiments.

9. Presentation of the semester project thesis - control.

10. Pruning - ANN optimisation. Semester projects - consultations.

11. Experiments with neural network parameters. Semester projects - consultations.

12. Hardware implementation of the Kohonen Self-Organizing Maps by FPGA

13. Semester projects - consultations.

14. Semester projects - evaluation, credits.

Study Objective:

Study materials:

1. Kohonen,T.: Self-Organizing Maps. Berlin Heidelberg, 3rd Edition, Springer Series in Information Sciences, Springer-Verlag, 2001, ISBN 3-540-67921-9.

2. Handbook of Neural Network Signal Processing.The Electrical Engineering and Applied Signal Processing Series. Ed.: Yu Hen Hu, Jenq-Neng Hwang. CRC Press, USA,2002, ISBN 0-8493-2359-2.

3. Haykin, S.: Neural Networks. A Comprehensive Foundation. Macmillan College Publishing Company, Inc. USA, 1994. 2nd.ed. 1998, Prentice/Hall, Upper Saddle River, NJ.

4. Program library SOM Toolbox 2.0. www.cis.hut.fi/projects/somtoolbox/download

Note:

Further information:

No time-table has been prepared for this course

The course is a part of the following study plans: